Assessing Heterogeneity of Correlation Matrices in Misspecified Meta-Analytic Structural Equation Models

Abstract

Meta-analytic structural equation modeling (MASEM) techniques are increasingly common tools to synthesize data across multiple studies. One popular approach is two-step MASEM, where study correlation matrices are pooled in a first stage using either a fixed- or random-effects model, to then fit one or multiple structural equation models onto the pooled correlation matrix in a second stage. In a simulation study, we examined the performance of different fit criteria and resulting parameter estimates under both random- and fixed-effects pooling when fitting a three-factor CFA model to study populations that were partly misspecified. We discuss benefits and issues when using a random-effects model in this scenario and discuss future research directions regarding correlation matrix heterogeneity when using MASEM methods.